Skip to content

Latest commit

 

History

History
92 lines (63 loc) · 3.77 KB

INSIGHTS.md

File metadata and controls

92 lines (63 loc) · 3.77 KB

Automotive Assembly Intelligence - Report Findings

Three years of plant telemetry (2023-2025): 35.3M units, 26,304 production hours across 3 lines, ~726k defects, ~8.1k equipment faults. Every figure below traces to a named SQL view over data that passes the integrity checks on the Methodology panel. Fully synthetic, seeded, no proprietary data.

This report is built to answer the questions automotive plant management asks, not to display vanity metrics. Each finding maps to a decision.


Executive: OEE and where output is lost

Plant OEE = 90.3% (Availability 93.0% x Performance 99.2% x Quality 97.9%), consistent across all three lines. OEE is shown with its three loss buckets and the formula, so a reviewer can see exactly how it is derived (v_oee, v_oee_by_line).

Where output is lost (no dollars - downtime hours + scrap units): Final Assembly and Paint are the largest combined loss sources; Spot Weld is the single biggest scrap origin. Final Inspection shows high equipment downtime but near-zero scrap origin - it catches defects, it does not create them, so corrective spend belongs on the process stations upstream (v_loss_by_station).

Decision: fix Paint and Final Assembly first; do not chase the inspection gate.


The invisible night shift

D-crew (nights) takes ~30% longer to clear the same faults than the best day crew, concentrated in the pre-handoff window. Same equipment, same fault mix - the gap is repair speed and coverage, and it never appears in daily output totals (v_mttr_by_crew, v_shift_handoff_effect).

Decision: a staffing/coverage problem, not an equipment problem.


Defect origin vs detection

~69% of defects are caught downstream of where they were created. The top three process stations (Paint, Final Assembly, Spot Weld) create ~63% of all scrap, but most of it surfaces at Final Inspection (v_rootcause_ranking, v_propagation, v_detection_ranking).

Decision: target quality effort at origin stations, not the detection gate.


Did our fixes actually work? (event rediscovery)

The quality trend independently surfaces real operational events straight from the fact tables, each matching the ground-truth log:

Event (date) Recovered from data
Weld-cell retooling (Apr 2024) Spot-weld defects step down ~43%, permanently
Supplier fastener bad batch (Oct 2024) Final-assembly defects spike, then recover
New model-year launch (Jan 2025) Yield dip + recovery on Trim/Final Assembly
Paint-booth upgrade (Sep 2025) Paint defects step down
PM / reliability program (Sep 2024) Mechanical faults down ~22% (6 months after vs before)

Decision: proof a process change moved the number - a step-change, not noise.


Reliability: capital plan

Per-robot fault counts with a 2024-to-2025 trend flag produce a replace-or-overhaul shortlist; several top contributors show a rising fault rate (v_robot_candidates). Thermal faults track summer severity (2025 the worst heat year), separating a seasonal pattern from a real regression (v_summer_thermal).

Decision: schedule the rising-trend robots for overhaul before next year.


Trust: methodology & validation

A dedicated panel shows row counts, referential-integrity checks (orphan faults, orphan shift links), yield reconciliation, null checks, the date range, and the provenance + seed (v_validation). All integrity checks pass. This is the "where did this come from and why can you trust it" section desk dashboards omit.


Sections in the live report

Executive (OEE + KPIs) · Where output is lost · Shift analysis · Defect origin vs detection · Event rediscovery · Reliability & capital plan · Methodology & validation. Each carries a plain-English "what this means / what to do" read.